To address the feature drift induced by inter-individual variability in marine main engines and the scarcity of fault data in the target domain, this study leverages digital twin technology and transfer learning to achieve cross-individual fault diagnosis.

Based on marine main engine performance parameters and an adaptive mechanism, a digital twin-enhanced cross-individual fault diagnosis model (DT-DANN) is developed. Particle swarm optimization (PSO) is employed to adaptively tune performance parameters factors, aiming to construct a high-fidelity individual digital twin model and generate healthy-state data samples under variable operating conditions. By integrating a multi-scale one-dimensional convolutional neural network (1D-CNN) with an improved domain adversarial neural network (DANN), a healthy-state anchor strategy is proposed to enable cross-individual fault diagnosis for marine diesel engines.

Simulation results demonstrate that the adaptively tuned performance parameter factors achieve a MAPE of 0.052 0%, while the proposed DT-DANN model reaches a diagnostic accuracy of 100%.

The proposed method effectively mitigates the model mismatch problem caused by inter-individual variability and data scarcity, enabling high-accuracy zero-shot cross-individual fault diagnosis for marine main engines.

In order to monitor the health status of a centrifugal pump in real time, this study proposes a model for the real-time identification of the degradation state of centrifugal pumps.

First, based on the operating parameters and degradation mechanism of the centrifugal pump, a combined weighting model using a combination of subjective and objective weights is used to calculate the combined weights, then a health index during the degradation process of the centrifugal pump is constructed. Second, based on the existing pump degradation data, a degradation identification model based on the genetic algorithm-group method of data handling (GA-GMDH) algorithm is proposed.

The reliability of the GA-GMDH monitoring model is relatively high, with a root mean square error of 0.029216 between the output values of the health index and the actual values. Based on the model's output results, the accuracy of degradation state identification is 93.333%.

The results of this study can provide valuable references for the health monitoring and maintenance operation management of centrifugal pumps.